KR20000071096A - A system for predicting future health - Google Patents

Abstract

The present invention provides a computer-based system for predicting an individual's future health, wherein (a) a subpopulation D of a member is identified as having acquired a specific biological condition within a specific time period or age interval, and the subpopulation of the member A computer comprising a processor comprising a database of time-series obtained biomarker values from individual members of a test population, wherein s is identified as not obtaining specific biological conditions within a particular time period or age interval; (b) (1) subpopulations D and Selecting a subset of biomarkers for discrimination between members belonging to the group based on a distribution of biomarker values of individual members of the test population; (2) (i) a subpopulation PD having a certain high probability of acquiring a particular biological condition within a particular time period or age interval for a member of the test population or a predetermined biological condition within a particular time period or age interval. Subpopulation with low probability The selected biotechnology to develop a statistical process that can be used to classify as belonging to, or (ii) quantitatively estimate, for each member of the test population, a probability of obtaining a particular biological condition within a particular time period or age interval. A computer-based system is provided that includes a computer program comprising using a distribution of markers.

Description

A SYSTEM FOR PREDICTING FUTURE HEALTH}

Once you have enough confidence and are able to predict in time that future health problems will arise for an individual, you will have the opportunity to prevent future health problems rather than wait for the disease to occur and treat the symptoms. It would be desirable to have more. Currently, much of the medical research funding has been spent on improving the diagnosis and treatment of asylum, rather than finding preventive measures to reduce the risk of asylum long before the commonly observed symptoms of the disease become evident. Although the focus on treating diseases may not only cure post-diagnosis, but also elaborate on the technologies and methods developed to diagnose existing diseases, these advances may bring significant advances in the medical sciences. This will increase the cost of treatment. The increasing cost of care is a major financial issue not only for individuals but also for society as a whole, increasing the demand for finding ways to reduce medical costs.

Therefore, when it is known in advance that the risk of disease is high, and effective preventive measures can be taken, it is possible to realize a substantial reduction in the total medical cost not only for the benefit of the individual but also for society as a whole.

To date, two of the problems encountered when trying to determine or predict an individual's future health status are as follows. (a) First, the prediction is inaccurate because the prediction of an individual's future health is based on data from a relatively small number of study samples consisting of hundreds or thousands of subjects. (b) Second, these predictions require extrapolation of individuals from the mean (and other factors) of the study sample. This extrapolation poses a big question about the reliable estimation of risk within a particular individual, or in a group that has a high risk for a particular disease (hereafter simply referred to as a risk group). This is partly true because the statistical process commonly used is not for individual populations but for the average of the population.

In order to obtain quantitative predictions, an "person's future health condition" should consist of the occurrence of a specific event at a specific time period. There are two examples. (a) Onset of myocardial infarction within a five year period, and (b) Death of the following year. The prediction of such an event should be based on probability, i.e. probability.

In this configuration, two types of probabilities are important. The pre-event probability of an event is the probability of the event before the fact of occurrence or non-occurrence of the event. The post probability of an event refers to the event after the event is realized, that is, the probability of the event after the fact of occurrence or non-occurrence of the event. If the event has occurred, the post probability of the event is 1, and if the event has not occurred, the post probability of the event is 0. This distinction between pre- and post-probabilities is worth considering.

Prior probabilities of events occurring in consecutive years or at different times can be important information. Knowing the probability of an event can change its behavior, and the actions that a person takes may depend on the event's prior probability. This principle becomes very apparent when considering two extreme cases. It is almost certain that a person exhibits different behaviors (five different actions) in two scenarios, two scenarios where the probability of human death in the next year is (a) 0.9999 or (b) 0.0001.

The prior probability of an event depends on the information available when the probability is evaluated. To illustrate this point, consider the following hypothetical "game."

Healthy individuals are randomly selected from all US residents and followed for any period of time (one year). After a certain period of time (one year), the individual's life state (survival or death) will be identified. "Event" is "the person who died during that period (1 year)". After a certain period of time, an event with post-probabilities 1 and 0, respectively, occurs (death) or does not occur (survival). Prior to selecting an individual, US mortality statistics can be used to estimate the prior probability that an individual will die within that period of one year. This probability is calculated as p = d / N, where N is the total number of individuals in the risk group, where all individuals in the US population are alive at the beginning of the period, and d is the total in the risk group. The death toll. For example, the data for 1993 is approximately d = 2,268,000, N = 257,932,000, and the prior probability of the event is approximately p = 0.0088 [Vital Statistics, Annual Report for the Year 1993 (Provisional Statistics), Death by Microsoft Bookshelf 1995 Almanac. ", And data from the" Vital Statistics of the United States "published by the National Center for Health Statistics. In this game, the preliminary probability of the event is based on very little information that it would be a member of a risk group consisting of all those who survived and the US residents at the time of selection.

Additional information about the risk group that randomly selected the subject means additional information about the subject and the change in the prior probability of the event. For example, continuing the "game" based on 1993 data is as follows.

If the risk group was a US male group, that is, the subject was known to be male before the selection, then the prior probability of the event would be approximately p = 0.0093, which is roughly comparable to when no gender was known or specified. 6% higher.

If the risk group was a U.S. male group between 75 and 84 years old, that is, the subject was known to be a male between 75 and 84 years old prior to selection, the prior probability of the event is approximately p = 0.0772, and the age is unknown. It is roughly 8.3 times higher than when not specified or unspecified.

These examples illustrate the general theory that the prior probability of an event depends on the information available when the probability is evaluated. Most accurate estimates of prior probabilities are usually based on all available information.

Highly accurate estimates of prior probabilities do not guarantee specific results. In other words, the prior probability for a particular individual may not be very close to the posterior probability. Considering the extreme cases mentioned above, the probability of premature death of a person in the following year is 0.0001. Survival is likely to be very high, but it is not guaranteed. Of all the individuals in this "game", approximately 9,999 out of 10,000 will survive that year, with a post-probability of 0 (this value is close to a prior probability of 0.0001), with approximately 1 of 10,000 deaths and a post-probability of 1 Which is very different from the prior probability value. To make this principle clearer, consider a fair coin tossing (back and forth decision) with exactly 0.5 prior probability of "front". The posterior probability of "before" is 0 or 1, and neither is very close to 0.5. Thus, the prior probabilities for an individual need not consider an approximation of the posterior probabilities for that individual. However, if a very large number of individuals are participating in the game, then the average of the posterior probabilities, the percentage of individuals in which the event occurs, will be very close to the prior probabilities.

In some cases, prior probabilities can be altered by "moving" into groups with different prior probabilities for an individual. For example, middle-aged men who are US residents with high total cholesterol levels and very low levels of lipoproteins are found to have a higher probability of dying from myocardial infarction over the next five years than those with very low cholesterol levels. lost. According to clinical studies, if an individual with a high cholesterol level can reduce his cholesterol level by moving to a group with a lower cholesterol level, he can reduce the likelihood of death from myocardial infarction over the next five years.

The term "risk", which is used in the following, will be used instead of "predetermined probability of a particular event in a particular time zone". This corresponds to the statistical definition of risk as expected death, where the death function has a value of 1 if an event occurs and 0 if no event occurs.

The above discussion shows the principle that if the levels of information are different, the prior probabilities are different. The risks for many known individuals (ie, members of small subpopulations with many known features) can be very different from those for large subpopulations with few known features. However, there are other problems that make it difficult for a population's usual scientific research to determine the risk of disease to an individual. This problem usually arises from too simple perception of the cause of the disease, in particular the cause of chronic degenerative diseases such as cancer, cardiovascular disease, diabetes and the like. That is, for a variety of causes, such a disease tends to be believed to be suppressed or clinically directed to treatment by prescribing a single component or a single drug compound. For example, breast cancer can be controlled by appropriately reducing fat intake, colon cancer can be controlled by adding fiber of certain foods, heart disease is clinically indicated by high blood cholesterol levels, and gastric cancer is low vitamin C. It has been known that it can be clinically indicated by blood concentration. These too simplistic views have been so often verified that they are inadequate for identifying causes for individuals. There are many confusing variables to consider, not to mention the great difficulty of extrapolating population data for individuals within a population. Examining and investigating a single component out of thousands, but not thousands, of possible component causes can be very uncertain, especially when attempting to extrapolate these data to estimate disease risk for an individual.

These two difficulties: (a) extrapolating the population data of the individuals from the experiment to randomly selected individuals, and (b) relying on a single treatment indication or cause of disease outbreak were randomly selected individuals. Seriously impair the estimation of future disease risks. If an individual's risk for a particular disease can be judged more reliably, it will be possible to provide information to those individuals who can make more informed decisions about their personal behavior. More reliable ways of predicting future health conditions must be a powerful means for individuals to manage their own health by making their own health status their own.

Moreover, for individuals whose categories may belong to several categories that have a high correlation with a particular disease, such as heart disease, and have been identified as having a high risk for a particular disease, currently available methods suggest that this disease may affect a particular individual. It is common to be quantitatively predicted with sufficient confidence when motivated to be lethal or motivated to motivate the individual, and to take effective measures in advance to reduce the risk. Therefore, there will be a need for a valid universal means that will not only reliably predict the occurrence of a particular health problem within a certain time period, but will also be useful in managing preventive measures taken based on this prediction.

The present invention relates to a computer system and method for predicting a future health condition of an individual. More specifically, the present invention obtains a set of time series data for a large number of biomarkers from a large human test population, statistically selects predictive biomarkers, and selects appropriate multivariables based on the selected biomarkers. It is intended to predict an individual's future health status by determining and determining an evaluation function.

1 shows population D (solid line) and population of an embodiment A diagram showing an experimental distribution function ("EDF") of linear discriminant function values based on the estimation for (dotted line).

2 shows population D and population of an embodiment Plots an experimental distribution function ("EDF") of a linear discriminant function based on a minimum random subject effect prediction value for.

An object of the present invention is to provide a means that can determine the risk of a disease in the future to bring a greater effect of prevention than the treatment of the disease.

More specifically, the present invention aims to provide a universal means for quantitatively predicting the risk of a wide range of diseases for a selected individual with substantially higher probability of future and greater reliability than is currently possible.

In particular, the present invention provides a computer-based method and apparatus that provides a system for determining future health risks for a particular individual and for managing preventive measures taken to reduce future health risks for a particular individual. There is a purpose.

The present invention identifies selected biomarker sets that include information about the probability that an individual will have a particular biological condition within a particular time period or age interval, and cross-sectional views of these biomarkers to estimate the individual's risk. ) And longitudinal values.

In addition, the present invention is a computer-based system for predicting the future health of the individual,

(a) a subpopulation D of a member is identified as having acquired a particular biological condition within a particular time period or age interval, and the subpopulation of said member A computer comprising a processor comprising a database of time-series obtained biomarker values from individual members of a test population, wherein s is identified as not obtaining specific biological conditions within a particular time period or age interval;

(b) (1) subpopulations D and Selecting a subset of biomarkers for discrimination between members belonging to the group based on a distribution of biomarker values of individual members of the test population;

(2) (i) a subpopulation PD having a certain high probability of acquiring a particular biological condition within a particular time period or age interval for a member of the test population or a predetermined biological condition within a particular time period or age interval. Subpopulation with low probability Classified as belonging to, or

(ii) quantitatively estimate the probability of acquiring a specific biological condition within a specific time period or age interval for each member of the test population;

A computer-based system is provided that includes a computer program comprising using the distribution of the selected biomarker to develop a statistical process that can be used to.

In addition, the present invention is a computer-based system for predicting the future health of the individual,

(a) a computer having a processor containing a plurality of biomarker values from an individual;

(b) collecting from the individual a plurality of biomarker values from which at least one biomarker value is obtained by physically measuring the biomarker value;

(i) the individual as a person with a certain high probability of acquiring a specific biological condition within a particular time period or age interval, or as a person with a certain low probability of acquiring a specific biological condition within a specific time period or age interval. Or (ii) applying a statistical process to the plurality of biomarker values to quantitatively estimate the probability of obtaining a particular biological condition within a particular time period or age interval for the individual. Computer program,

The statistical process is,

(1) the subpopulation D of the member is identified as having acquired a particular biological condition within a particular time period or age interval, and the subpopulation of the member Collecting a database of time-series obtained biomarker values from individual members of the test population, wherein is identified as not obtaining specific biological conditions within a particular time period or age interval;

(2) subpopulation D and from the biomarker Selecting a subset of biomarkers for discrimination between members belonging to the group based on a distribution of biomarker values of individual members of the test population;

(3) A computer-based system, characterized in that based on using the distribution of the selected biomarker to develop the statistical process.

Preferred embodiments of the present invention are described in detail, and the embodiments of the present invention are for illustration and are not limited to this embodiment.

The present invention relates to a condition in which a person's health is tentatively showing nutritional, toxicological, genetic, hormonal, viral, infectious, anthropometric, lifestyle, and abnormal physiological and putative pathological conditions of the individual. It is based on the theory that it is affected by the complex interaction of a wide range of physiological and biochemical factors involved. In accordance with this theory, the present invention provides a quantitative prediction of an individual's future health based on a statistical comparison of an individual's biomarker values with a time series obtained database of a set of multiple personal biomarker values for a large test population. It aims to provide a practical system for predicting future health using multivariate statistical analysis techniques that can be provided. As used herein, the term "biomarker" refers to a biological indicator that may be associated with or affect the diagnosis or prediction of personal health. In addition, the term "time series" means that the biomarker value is obtained periodically for a time period, in particular for at least two measurements.

The frequency and duration of time series estimates may change. For example, some biomarkers may be judged on a yearly basis for periods as short as two years to as long as lifetime. Under certain circumstances, such as the estimation of newborns, biomarkers may be assessed more frequently, for example daily, weekly or monthly. In the case of time series estimates, the period can be randomly specified, that is, occur at unequal time intervals. The set of time series estimates for an individual can be “complete”, and complete means that data from all planned estimates and all planned biomarkers can actually be used and used. In addition, a set of time series assessments of an individual may become “incomplete”, meaning that in some cases the data is not complete. The biomarker of an individual can be assessed at specific time or time series. The present invention can carry out a statistical analysis of the necessary data obtained from an individual having any or all of the above mentioned features, i.e., a particular temporal or time series, regular or irregular, complete or incomplete for a time.

The system of the present invention for assessing future health provides a quantitative estimate of an individual's probability of having a particular biological state within a certain time period. This quantitative probability estimate is calculated using the sequential statistical analysis of the present invention. The system of the present invention will typically be used to provide quantitative predictions of future biological conditions for the future of 1, 2, 3, 5 or even more than 15-20 years. While the system of the present invention is generally used long before any signs of a particular disease are observed or examined, the system of the present invention can be used to predict future health over a relatively short time period of just months, weeks or even shorter periods. It may be.

Although there is no upper limit on the number of members that can be included in a test population, it may include millions of test members, but a representative test population will initially contain a very small number of members. The test population can be selected from a large general population using appropriate statistical sampling techniques to improve the reliability of the collected data.

In a preferred embodiment, the invention provides a set of mathematical and statistical functions for estimating the risk status of individuals with a particular biological condition within a particular time period or age interval, and which can be used to identify the most at risk individuals. A computer system using statistical analysis steps is provided. Prior to step I of the method of the present invention, the available subjects may be randomly assigned to a training sample or an evaluation sample. Steps I-III are for data from training samples and Step IV is for data from evaluation samples. Stage I is a screening step using correlation analysis, logistic regression, mixed model analysis and other methods to select a large subset of biomarkers with potentially useful information for risk estimation.

Step II is a factor estimation step that uses a mixed linear model to estimate the expected vector of the candidate biomarker and the structured covariance matrix factor, even in the presence of incomplete data and / or irregularly time-defined time series data. Step III selects informative biomarkers (including relevant time series evaluations), estimates discriminant function coefficients, and uses discriminant analysis and logistic regression to use inverse cumulative distribution functions and logistic regression to estimate individual risks. Biomarker selection and risk assessment. Step IV is an evaluation step using an evaluation sample to provide unbiased estimates for the misclassification rate of the determination process.

Although the individual steps of the aforementioned statistical process are disclosed in the statistical literature, these individual steps have never been combined in a single overall process as disclosed herein. In particular, the following conventional methods are described, for example, by "Encyclopedia of Statistical Sciences" by Samuel Kotz, Normal L. Johnson, Campbell B. Read, published by John Wiley & Sons in 1985. "(A) Correlation Analysis (Vol. 2, pp. 193-204), (b) Logistic Regression (Vol. 5, pp. 128-133), and (c) Mixed Model Analysis (Vol. 3, 137). Pp. 141, titled “Fixed-, Random-, and Mixed-Effect Model”), (d) Discriminant Analysis (Vol. 2, pp. 389-397).

The present invention can use the conventional methods of these processes, their methods of improvement and newer methods of these processes that will sometimes be improved and known.

Correlation analysis is a term used in statistical methods to estimate the degree of relevance of a linear relationship between two or more variables. Correlation as used herein includes various types of correlations, including Poisson product-moment correlation, Spearman's ρ, Kendall's τ, Fischer-Yates' γ _F and It is not limited to anything else.

Logistic regression is a term used in statistical methods used to analyze the relationship between tested dependent variables (which can be either proportional or rate) and analytical variables, including linear algebraic-linear models. The application of logistic regression as used herein is primarily used for analysis in which the dependent variable is a binary output representing an individual in one of two complementary (non-overlapping) populations of subsectors. These two groups are those that will have a particular disease or condition at a particular time period or age (sometimes referred to herein as “specific biological conditions”) and those that will not have a particular disease or condition at a particular time period or age. The interpretation variable here is typically a function of the biomarker or biomarker.

Mixed model analysis involves correlated dependent variables (multivariate measurements or observations, time series measurements or observations of one variable, and / or time series multivariate measurements or observations), covariances such as age, and classification variables (representing members of a population). The term is used in the analysis of expected value relationships between “independent variables” which may include and is used in statistical methods used in the analysis of structures and factors representing covariances between correlation measurements or observations. "Mixed model" is a term that includes fixed models, random models, and mixed models. The mixed model may have a linear or nonlinear structure in the expectation model and / or covariance model. Mixed model analysis typically includes an expectation factor (sometimes denoted by β) and a covariate matrix factor (sometimes Σ = ΖΔΖ '+ V, where Δ and V are unknown matrixes). . Mixed model analysis may also include random subject types (denoted as d _k for the _kth subject) and so-called "optimal linear fleet predictors (or" BLUPs ") for individual subjects. It is common to include procedures for verifying hypotheses about factors and / or covariance factors, and constructing confidence domains for factors.

In particular, a discriminant assay is one of two complementary (non-overlapping) groups of subjects (eg, a group that will have a particular disease or condition at a particular time period or age, and a group that will not have a particular disease or condition at a particular time period or age group). In relation to a statistical analysis that improves a discriminant function that can be used to assign multivariate observations (eg, vector values of biomarkers from one subject) according to that value. In addition, the discriminant function can be said to be the function used as the basis for calculating an estimate of the probability that a given observation belongs to a given group. For the present invention, the associated observations typically have a plurality of biomarker values obtained from each member of a large test population or from individual validation subjects. The discriminant function of the present invention utilizes the distribution of these biomarkers with which each determined biomarker is associated. This distribution consists of the total number of individual members of the test population with each biomarker value and the biomarker value itself. Accordingly, the present invention provides a statistical approach that utilizes individual biomarker values obtained for each biomarker of an individual member from a test population, and separately from, for example, a distribution based on mean biomarker values obtained from different test populations for different biomarkers. Adopt the process.

"Discrimination function" includes, but is not limited to, classification of observations (scalars or vectors) into two or more groups, such as, for example, linear discriminant functions, quadratic discriminant functions, nonlinear discriminant functions, and various types of so-called optimal discriminant processes. The term means any one of several different types of functions or processes.

The computer system of the present invention can drive a computer program or a series of computer programs (hereinafter simply referred to as "computer programs"), and includes steps for performing necessary operations and data processing in various steps and steps of the present invention. A computer having a processor. This processor can be a microprocessor, a personal computer, a mainframe computer, or any digital computer capable of running a computer program that can generally perform the necessary computation and data processing. The processor includes a central processing unit, random access memory (RAM), read-only memory (ROM), one or more buses or channels for transferring data between multiple devices, one or more display devices (e.g. monitors), one or more input / output It is common to have devices (eg, floppy disk drives, fixed hard disk drives, printers, etc.), and adapters to control input / output devices and / or display devices and to connect these devices to a bus or channel. Some particular processors may include all of these devices, or some of these devices only.

The computer program may be stored on a ROM, a disk or a set of disks, and other accessible media that can be used to store and distribute the computer program.

The computer program can perform operations for various steps and steps of the method on specific temporal and / or time series multivariate biomarker data.

Biomarker data is preferably collected from a sufficiently large test population to ensure that the total number of members with a particular relevant biological condition within a two to three year cycle is significant for the particular biological condition and is large enough for the adopted discriminant assay. . One of the features of the present invention is a means of using the same database for predicting death with any major disease and / or death due to major underlying causes of death in short cycles such as one to two years. As such, the test population is large enough to apply the subject system to any of the more common diseases and the underlying cause of death, which is at least about 60%, preferably at least about 75% of the total deaths involved. It is preferable. The relevant deaths here refer to those who have died pathologically, as distinct from accidents, murders and suicides.

For example, using data obtained from the Monthly Vital Statistics Report, February 26, 1996, Supplement, Volume 44, Center for Disease Control and Prevention, approximately 75% of all pathologically deceased deaths were approximately 880 Malignant tumor with a net mortality rate of 205.6 / 100,000 (ICD 140-208), major cardiovascular disease with a net mortality rate of 367.3 / 100,000 (ICD 390- 448), chronic lung disease with net mortality of 39.2 / 100,000 (ICD 490-496), and death of diabetes with net mortality of 20.9 / 100,000. These diseases are those that actually respond to altered diet and lifestyle conditions and exhibit major dietary and lifestyle impacts represented by various definable and measurable biomarkers.

As one of the inherent features of the present invention, computer systems and devices can be used to identify individuals with any of these major diseases based on comparing individual profiles of biomarker values to biomarker values obtained from members of large test populations. Can be used to determine the risk of Since these major diseases are known to share many common factors that can be reflected in biomarker values, subject computer systems can be used to simultaneously assess the risk of having these major diseases. For example, total serum cholesterol is known to be a biomarker associated with many of these diseases. Each subject of a biomarker, which is an important predictor, is administered in combination with other important biomarker predictors of a particular disease, or as the basis for the cause of death, and by using the present invention to compare this profile with a test population, an individual subject May have a certain quantitative reliability and be informed that the disease is most dangerous for a particular individual.

A feature of the present invention is that in individuals who are most at risk of having a particular disease, a quantitative probability of having the disease within a certain time period or age interval may be provided in the future before the usual symptoms of the disease become apparent. When information is provided about a number of known diseases in response to altered regular expressions and lifestyle conditions, individuals may make behavioral changes that may reduce the risk of identified diseases.

In addition, as more data is provided for a larger number of subjects over a longer period of time, it is seen that the basis for more detailed classification and cause of death for each of the major and less common causes of disease is seen. It may be defined and included by the method of the invention. For example, classification may be made into various types of cancers such as liver cancer, lung cancer, gastric cancer, prostate cancer, and the like. The computer system of the present invention thus comprises a larger portion of the population to predict the quantitative risk of each individual with or without a specified pathologically deduced disease within a certain time period where the disease is defined by successively narrower characteristics. Provide means.

The comprehensive biomarker set from which biomarker data is collected from the test population preferably includes as many diverse biomarkers as known or believed to be related to the underlying cause of the most common disease and pathologically deduced death. In addition, the values of representative biomarker populations from each of the known and generally accepted genetic, physiological and biochemical regions of biological function can be included. For example, additional biomarkers are collected that are preferably included for all things that can be measured in biological samples that can be stored for analysis after sampling has been made.

The biological sample preferably includes blood vessel and urine samples, but other biological samples may be included. For example, samples of saliva, hair, toenails and nails, face, exhaled air, and the like can also be collected. Such biological samples are typically obtained from virtually all members of the test population. However, in some cases, a subset of specific biomarkers can only be obtained from a particular subset of the population.

Biomarker data related to nutritional habits and lifestyles as well as biological samples are collected from each member of the test population. Biomarkers related to nutritional habits and lifestyles may include, for example, those shown in Table 1. The nutritional and lifestyle buyer markers listed in Table 1 represent biomarkers of the type related to nutritional habits and lifestyles, but do not exclude biomarkers related to nutrition and lifestyles within the scope of the present invention. Biomarkers representing clinical and epidemic biomarkers as well as important nutritional judgments can be determined by other factors, such as nutritional intake. Thus, the category descriptions shown in Table 9 only represent exemplary classifications of categories that can be selected to obtain biomarker values. Nutritional and lifestyle biomarkers, which may change over time, are preferably collected and recorded for each member of the test population each time a biological sample is collected.

Table 1

A list of biomarkers that can be used in the method of the present invention for predicting future health status is described.

Serum Biomarkers

Total cholesterol *

HDL Cholesterol *

LDL Cholesterol *

Apolipoprotein b *

Apolipoprotein A ₁ *

Triglycerides *

Lipid peroxide (malondialdehyde equivalent: TBA) *

α-carotene (collected with lipoprotein carrier) *

β-carotene (collected with lipoprotein carrier) *

γ-carotene (collected with lipoprotein carrier) *

Zeta-carotene (collected with lipoprotein carrier) *

α-cryptoxanthin (collected with lipoprotein carrier) *

Lycopene (collected with lipoprotein carrier) *

Rutin (collected with lipoprotein carrier) *

Anhydro-Routine (collected with lipoprotein carrier) *

Neurosporene (collected with lipoprotein carrier) *

Phytofluene (collected with lipoprotein carrier) *

Phytoenes (collected as lipoprotein carriers) *

α-tocopherol (collected with lipoprotein carrier) *

γ-tocopherol (collected with lipoprotein carrier) *

Retinol *

Retinol Binding Protein *

Ascorbic acid *

Fe *

K *

Mg *

Full Inn *

Weapon Inn *

Se *

Zn *

Ferritin *

Full iron bonding capacity *

Fasting Glucose *

Urea Nitrogen *

Elemental Acid *

Prialbumin *

albumin*

Whole protein *

Bilirubin *

Thyroid Stimulating Hormone T3 *

Thyroid Stimulating Hormone T4 *

Cotinine

Aflatoxin-albumin product

Hepatitis B Anti-Core Antibody (HbcAb)

Hepatitis B surface antigen (GhsAg +)

Candida albicans antibody

Apstein-Barr virus antibodies

Type 2 herpes simplex antibodies

Human Papilloma Virus Antibodies

Heliocobacter phyllori antibody

Estradiol (E2) (adjusted for female cycle) *

Sex hormone binding globulin *

Prolactin (adjusted for female cycle) *

Testosterone (adjusted for female cycle) *

hemoglobin*

Mystic Acid (14: 0) *

Palmitic acid (16: 0) *

Stearic acid (18: 0) *

Arachidic acid (20; 0) *

Behenic Acid (22: 0) *

Tetracos-A1 Ixane (24: 0) *

Mystic Oleic Acid (14: 1) *

Palmitic Acid (16: 1) *

Oleic acid (18: 1n9) *

Guardoleic acid (20: 1) *

Acidic acid (22: 1 n9) *

Tetracosenoic acid (24: 1) *

Linoleic (18: 2n6) *

Linoleic acid (18: 2n6) *

γ-gamma linoleic acid (18: 3n6) *

Eicos adienoic acid (20: 2n6) *

Di-homo-γ linoleic acid (20: 3n6) *

Arachidonic acid (20: 4 n6) *

Eicospentaenoic acid (20: 5n3) *

Docosatetraenoic acid (20: 4n6) *

Docosapentaenoic acid (20: 5n3) *

Docosahexaenolic acid (20: 6n3) *

Totally unsaturated fatty acids (16: 0, 18: 0, 20: 0, 22: 0, 24: 0) *

Total monounsaturated fatty acids (14: 1, 16: 1, 18: 1n9, 20: 1, 24: 1) *

Total n3 polyunsaturated fatty acids (18: 3n3, 20: 5n3, 22: 5n3, 26: 6n3) *

Total n6 polyunsaturated fatty acids (18: 3n6, 20: 2n6, 20: 3n6, 20: 4n6, 22: 4n6) *

Total n3 polyunsaturated / total n6 polyunsaturated fatty acids (18: 3n3, 20: 5n3, 22: 5n3, 22: 6n3 / 18: 3n6, 20: 2n6, 20: 3n6, 20: 4n6, 22: 4n6) *

Total polyunsaturated fatty acids (18: 2n6, 18: 3n3, 18: 3n6, 20: 2n6, 20: 3n6, 20: 4n6, 20: 5n3, 22: 4n6, 22: 5n3, 22: 6n3) *

Total polyunsaturated / saturated fatty acids (18: 2n6, 18: 3n3, 18: 3n3, 20: 2n6, 20: 3n6, 20: 4n6, 20: 5n3, 22: 4n6, 22: 5n3, 22: 6n3 / 16: 0 , 18: 0, 20: 0, 22: 0, 24: 0)

[Approximately 10-30 genetic markers depending on the disease investigated]

Urine Biomarkers

Orotidine

Cl *

Mg *

Na *

Krytinine

volume

NO ₃

Aflatoxin (AF) M ₁

AF N ⁷ guanine

AF P ₁

AF Q ₁

Aplatoxycol

8-deoxy guanosine

Nutritional intake from food (by question document)

Whole protein *

Animal Proteins *

Vegetable Protein *

Fish Protein *

Geology *

Soluble Carbohydrate *

Whole Dietary Fiber *

Total calorie *

Percentage of calorie intake from lipids *

cholesterol*

Ca *

P *

Fe *

K *

Mg *

Mn *

Na *

Se *

Zn *

Total Tocopherols (collected from lipid intake) *

Before Retinoids *

All carotenoids *

Thiamine *

Riboprabin *

Niacin *

Vitamin C *

[About 30 different types of foods] *

[About 30 different fatty acids] *

Red blood cells

RBC Glutathione Reductase *

RBC Cataras *

RBC Superoxide Dismutas *

Anthropometric factors

kidney*

weight*

* Indicates biomarkers indicating important nutritional determinants.

Biological samples are analyzed to determine biomarker values for each component in the biological sample for which biomarker values are desired. It will be appreciated that all components found and measured in biological samples are within the scope of the present invention. For example, genetic biomarkers that can be measured in blood samples and biomarkers that can be measured in any other suitable biological sample can also be included.

Since another feature of the present invention is the identification of new biomarker sets useful for predicting disease and death, biomarkers that have not been previously known to have statistical significance for predicting specific diseases or causes of certain deaths. It may include a marker. Thus, since the total number of biomarkers that can be used is generally not limited, the number of actual biomarkers that can be used is generally only limited by actual boundary and methodological considerations.

Since another feature of the present invention is to provide a computer system for predicting a specific biological condition in a particular time period or age interval in the future, the total number of biomarker values is statistically important for predicting a single specific biological condition. Only biomarker values with Thus, although most of the subject systems are believed to be used as a universal means for predicting and managing the underlying causes of all major types of disease and death, the use of the methods disclosed herein is intended for one disease or death cause at a time. It may be.

Biological samples can be collected and analyzed immediately, and samples can also be stored after analysis. Samples are preferably stored after analysis because it is expected that a large number of samples can be collected for a relatively short time period and under circumstances that do not benefit the immediate practical analysis. Since the samples are generally stored for a substantial period of time, the samples are typically frozen.

These samples are stored and transported using conditions that preserve the sample completely. Such techniques include, for example, Diet, life-style and mortality by Chen, J., Campbell TC, Li, J., Peto, R of China, Oxford, England, Ithaca, 1990 , A Study of the Characteristics of 65 Chinese Counties, PRC Beijing, Oxford University Press, Cornell University Press, People's Medical Press.

Using physical samples, such as biological samples, is particularly desirable because these samples provide a practical means of providing abundant data of time-series acquired biomarkers that can be collected, stored, and analyzed using established cost-effective techniques. Do. Biological samples should be collected for a test population over an extended period of at least 5-10 years, most preferably 15-20 years or more, so that the quality of the data generated will continue to provide more reliable probability predictions.

Since the reliability of the system of the present invention is finally determined by the quality of the collected biomarker data, appropriate measures are taken to ensure the integrity of the data in all respects. For example, taking into account the stability of biomarkers, it is necessary to take appropriate measurements to account for many factors that can cause the effects or deterioration of biomarker values over time.

In addition, it is common for subjects to be described to obtain biomarker data from physical samples obtained from members of the test population or from a validation subject, and from biomarker data from the regular expression and lifestyle tests of each test, but biodata obtained from any data source. The use of marker data is within the scope of the present invention. For example, the methods of the present invention can be used for medical diagnostic data obtained from electronic biological measurement techniques such as electroencephalogram (EEG) data, electrocardiogram (ECG) data, radiation (X-ray) data, magnetic resonance imaging (MRI), or biological samples and regular expressions. And using only biomarker data obtained in time series from a lifestyle test or preferably a combination thereof.

Since the test population is preferably administered on a yearly basis, it can be expected that mortality will be tested for the test population representing the entire general population. For each mortality rate in the test population, individuals are identified and the underlying cause of death is recorded and it is desirable to use known coding systems. An example of a known coding system is the International Statistical Classification of Diseases and Related Health Problems (ICD-10), the tenth edition of the 1992 World Health Organization (WHO). Other encoding systems can also be used and fall within the scope of the present invention.

By using an efficient system for identification when a member of a test population has a disease or specific biological conditions, morbidity data is collected in addition to collecting the biomarkers and mortality of the test population.

The database of biomarker values records the incidence, medical conditions, medical pathology, or death of the disease, including information, diagnosis, and date of onset of each biomarker and biomarker samples from which the data and age were recorded when they were collected and recorded. It is desirable to include accurate information from an individual's examination. This database contains the values of the biomarkers evaluated before, during and after each onset, and this database is viable.

As one aspect of the present invention relates to the identification of biomarkers that are not yet known for their statistical significance for predicting future onset of the underlying cause of a particular disease or death, as many biomarkers as possible should be managed. In a preferred embodiment, approximately 200 biomarker values are obtained from each member of the test population, although without substantially limiting the number of biomarkers that can be used to develop computer-based statistical analysis methods.

Because the present invention aims to provide a practical and reliable system for predicting specific biological conditions at a particular time period or age, a substantially complete set of bios from each member of the test population at at least two different times. Markers are collected. In order to obtain information on trends and changes over time, it is more desirable that a complete set be collected at at least three different times, and more preferably that the biomarker values are collected in a substantially viable period.

In another aspect of the present invention, the present invention is based on the theory that the rate of change or change in the rate of change of an individual individual biomarker value of an individual may be more important for predicting future health status than the actual level of a given biomarker value. have. The discriminant function is typically determined using a substantially complete set of biomarker values. The practical reason why it is not reasonable to expect that a complete set of biomarker values as a whole is obtained from all members of the test population in all test cases is that the statistical analysis method of the present invention takes into account reliability for incomplete data in a statistically valid manner. Because you know how to include it.

Another object of the present invention is to provide a means for quantitatively evaluating the risk of a particular disease in the future and to provide a practical means for defining and identifying those biological conditions with the lowest risk among all future diseases. Thus, a "specific biological condition" is meant herein to encompass the full range of health conditions, from the healthiest health conditions to the most serious disease states. Accordingly, the present invention is to provide a system for managing and predicting future health from the healthiest state to the worst health conditions.

Although the results obtained from the test population can be used to predict the future health status of the general population in a particular region (country), it is not necessary to select a test population from the same general population for which individual future health predictions will be made. This restriction does not need to obtain a disease that is characteristic of the area in which they travel because it is known that a population of individuals having a probability of disease that is characteristic of their area and to a new area of the population that has a different set of probability of disease. This happens at the same time and in accordance with the acquisition of regular diets and lifestyles in the new area. That is, all races and ethnicities in the world tend to have the same general disease, regardless of their own characteristics, and can be unique to each race or ethnicity.

One of the features of the present invention is to provide a system provided to predict when future development of a disease problem will occur before the problem is usually diagnosed. The timing of future outbreaks of a particular health problem for a particular individual can be predicted by a specific quantitative probability estimate based on applying subject discrimination methods to a database collected from a large trial population. In addition, the present invention provides a system for predicting specific health problems in the future, and provides a system in which more data is collected with greater reliability for a larger test population over a longer period of time.

Biological samples are typically analyzed for each biomarker that requires quantitative values. For reasons of cost and convenience, with a large number of samples that can be collected, the samples are randomized for the rest of the test population as well as for individuals who have already been diagnosed with the disease or who have died during the time period in which the samples were collected. Only selected parts will be analyzed initially. For example, if the yearly mortality rate of the tested test population is typically in the range of 2-3%, a test population with 300,000 members will show a yearly mortality rate of 6,000-9,000 deaths, with the majority of deaths at each of the underlying causes of death. Will be due.

Another feature of the invention includes the step of waiting for a number of deaths to occur in a test population to select those whose biomarker values were initially determined. In addition, the population of currently valid validation members will be selected from the rest of the test population. Since we need to balance the need for enough samples with the need to adjust the cost to obtain statistically significant results, the system of the present invention only needs to provide a sample that will tend to provide a large amount of information for the lowest cost. Provides a way to limit the cost of analytical measurements. Inherently, the greater the number of deaths occurring in the test population, the greater the number of samples to be analyzed over time. However, the value of the data obtained from the point of view of establishing a more reliable quantitative prediction of future health status will be somewhat balanced with the cost of obtaining additional biomarker values. This is one of many features of the present invention that distinguishes it from other known conventional systems. This technique of delaying sample analysis allows a cost delay until the obtained result has a larger actual value.

When selecting a sample to be analyzed, biomarker values can be determined using methods known in the art. Since many samples are analyzed by separate measurements for multiple, if not most, biomarker values, these measurements are multi-channel analyzers such as Boehringen Mannheim Corp., Indianapolis, Indiana. It is generally made using the BMD / Hitachi Model 747-100 manufactured by. Such an analyzer can be designed to simultaneously measure biomarker values of selected multiple parts of the biomarker using a relatively small portion of the entire sample. For example, the amount of blood collected is typically approximately 15 ml, but only approximately 10-30 ml may be required for analytical measurements. Similarly, the amount of urine collected is approximately 50 ml, but approximately 100 μl is required for the analysis. Small amounts can also be used as appropriate for other biological samples.

In a preferred embodiment, any bioavailable detectable in a given sample after the sample has been collected since physically conservable biological samples can be used and relatively small amounts of analytical sample can be used for measurement at any randomly selected time. Using the marker, the method of the present invention can be effectively applied. For example, this system can be used to initially analyze what is currently considered the most important biomarker, but the system can easily be adapted to include other biomarkers that have not yet been recognized in order to have importance for predicting future health conditions. have. In general, with adequate time and economic data, all biomarkers detectable in preserved biological samples will be finally determined.

It is desirable to obtain a substantially complete set of biomarker values for each member of the test population, but it is common to realize that when samples are collected in time series from a geographically widely distributed population base, the realization becomes very difficult. . If an incomplete set of data is abandoned and not used at all using conventional statistical analysis methods, the amount of substantial data that ultimately occupies most of the initial test population will need to be abandoned. This can lead to substantial waste of data and serious qualitative degradation as a result of the rest of the data. This computer-based method includes a feature that provides a means of using all data collected by using statistically verifiable techniques to fill in "missing values." This is particularly useful for subject methods based on collecting an amount corresponding to a large amount of data compared to any prior art for a large number of validation members from geographically dispersed test populations. Obtaining comprehensive data from various large test populations is particularly desirable for obtaining biomarker values from members with a wide range of regular expressions and lifestyles representing entire human experiments.

The following terminology is described to describe the present invention.

For example, "specific biological conditions" refers to any one of the following descriptions.

Certain diseases (eg, diabetes) as classified, for example, in the International Statistical Classification of Diseases and Related Health Problems, supra;

Certain medical or health conditions or symptoms (eg, hypertension, limited to a deviation of the biomarker or biomarker value from a normal normal distribution);

Certain medical events and their sequelae (eg, ischemic seizures and resulting death or non-death and partial paralysis and related conditions associated with seizures; myocardial infarction and resulting death or non-death and MI-related conditions);

Premature death due to any cause (death at an earlier age than the average age at death, such as withdrawn from the sex and age of the individual in the initial assessment);

Death at certain ages;

A newly defined category based on having a particular set of biomarker values for a particular set of biomarkers.

Acquisition or onset of specific biological conditions refers to a condition in which an individual does not have a specific biological condition at a given evaluation time, but continuously tests a specific biological condition. Here, when an individual is mentioned to have had a specific biological condition, the onset is defined as what occurs when the individual has the specific biological condition.

For populations of people with or without specific biological conditions and specific biological conditions, population D and population There are two complementary subpopulations defined by

Population D: A subpopulation of individuals who will have specific biological conditions within certain time periods. Certain time zones may refer to specific almanac time periods (eg, "5 years to come"), specific age intervals (eg, "65 to 70 years") or similar time or age intervals.

group : Subpopulation of individuals who will not have specific biological conditions within a certain time.

A subset of these subjects is specified in part by specific time series pattern data for a large number (as many as possible) of biomarkers. Time series patterns include levels or structural concentrations of biomarkers and changes in levels. If one recognizes that a time series biomarker pattern partially specifies a subpopulation and has the data needed from a particular individual, that individual is group D or population It can be classified into one of two complementarity groups depending on which group they appear in.

Population PD: A group of people who are expected to have certain biological conditions at a particular time zone early in a particular time zone, ie, appear to belong to group D. These individuals are described as having a high probability of having certain biological conditions at certain times.

group : Prediction, that is, a group that does not have a specific biological condition at a specific time zone at the beginning of a specific time zone. A group of people that appear to belong to. These individuals are described as having a low probability of having certain biological conditions at certain times.

The term "expected high probability" means that having a low probability of several percent, perhaps less than 1%, or a probability of 10%, 20%, 50% or substantially higher percentages will vary depending on the particular biological condition. Can be. For example, the increased risk of lung cancer from smoking is important, even if the actual multiple risk increases from smoking are as long as 15-20 years or more, even if the probability of developing lung cancer is in the 5-10% range. Preferably it can be recognized as an inevitable risk. In some cases, for a particular biological condition to which the system is applied, a quantitatively determined probability can be determined. "Expected low probability" can simply be specified as a probability that does not belong to a high risk population with certain biological conditions, and the term can be individually specified by specific value.

Group D, or population, with a statistically appropriate number of members of the test population For each member of the test population with a specified biological condition within a particular time period or age interval, when identified as belonging to the group PD and Statistical process for estimating the probability or estimating the probability, ie the probability or group belonging to the group PD To determine the probability of belonging to a biomarker value of a member of population D, the population is determined using the method of the present invention. Can be compared with In a preferred embodiment of the invention, the member is associated with a group PD The statistical process to classify as will be in the form of a discriminant analysis process as described below, which may be referred to as a "discriminant procedure or discrimination procedure". A "statistically appropriately defined number" can be defined as the total number of biomarkers used in the analysis and the total number of verification members for which biomarker values are available are large such that convergence is obtained for the computational process used in the subject method. have.

The discrimination process has two related error rates.

(1) the probability of a false positive proposition, that is, classified as a group PD but the actual group The probability of a future subject to belong to.

(2) the probability of false negative propositions, ie groups The probability of future subjects classified as but belonging to the actual population D.

A preferred embodiment of the present invention will be a related method for obtaining accurate estimates of these two error rates.

A preferred embodiment of the present invention consists of three steps, each of which has a number of steps. The three steps are as follows:

Step Ⅰ. Set up evaluation methods and select biomarkers for consideration.

Step II. Estimation of candidate models with mixed discriminant, covariance structure, and predictive values.

Step III. A discriminant function operation using the estimated mean and the predicted value, and a logistic predictive value operation for each subject; Error rate estimation for the discriminant function.

Each step has several steps.

Some set of steps is repeated within a step. That is, a particular set of steps can be repeated many times until a particular purpose is achieved. Steps and preferred embodiments of these steps are described next.

Step Ⅰ. Set up evaluation methods and select biomarkers for consideration

The following steps will appear in the preferred embodiment of the present invention.

Step 1: Choose a method to estimate the error rate of the process

This method embodies any statistically appropriate method of estimating the error rate. Two of the many methods that can be used are a training sample / effective sample and a subsampling (or “resampling”) method.

Training sample / effective sample method; For the training sample / effective sample method, the test population is optionally divided into two subsets. These two subsets are "training samples" and "effective samples". All subjects (members of the test population) are assigned to either training samples or valid samples. The data from the subjects in the training sample is used in statistical analysis to derive specificity of the discriminant process and the probability estimation process. Data from the subject in the evaluation sample will be used to estimate the distribution of error rates and probability estimates of the discrimination process.

Subsampling method; "Subsampling" refers to a class of statistical methods, including jackknifing and bootstrapping, that can be used to produce a reduced estimate of error rate. In the subsampling method, data from all subjects is used in a statistical analysis that leads to the distribution of a particular and / or probability estimate of the discrimination process. By using all the data, in particular, when (1) the test population is not large, or (2) the prior probability of having biological conditions is small, even a large test population can be better discriminated than that obtained by the training sample / effective sample method. Process and / or probability estimation process. In the present invention, the subsampling method is intensive computation.

Step 2: Select "Training Sample", which is a subset of the test population used for statistical analysis leading to the Discrimination / Probability Estimation process, and "Valid Sample", which is a complementary subset.

If a subsampling method is used, the data from all subjects is used in a statistical analysis that leads to the distribution of a particular and / or probability estimate of the discrimination process. In this case, the "training sample" is the entire test population.

If a training sample / effective sample method is used, the training sample will contain approximately a certain percentage of the test population. In many cases, the training sample rate will account for 50%, although other rates may be used. The valid sample will include all subjects that are not included in the training sample.

Any subject assignment for training samples will typically be divided by subject age. Subject age is classified into appropriate intervals. In other words, the age group consists of all subjects whose age belongs to a specific age section. The interval is chosen so that the number of subjects in each floor is appropriate for statistical analysis. Within the age group, subjects will be randomly assigned to training samples or valid samples. Randomization is used to achieve a specific subject ratio in the training sample. For example, if the training sample is specified to contain 75% of the test population, approximately 75% of the subjects will be randomly assigned to the training samples within each age group. For example, if you specify one age group with ages 65 and below 70, 75% of the subjects in the floor will be randomly assigned to the training sample.

Step 3: Create a list of potential biomarkers that are possible discriminators

The purpose of this step is to list all suitable and potentially useful biomarkers called potential biomarkers. In a preferred embodiment, the list of potential biomarkers will include all recorded quantitative and personal characteristics of the subject in the test population. This list will include characteristics that do not change over time (eg, date of birth) and time-dependent characteristics such as body weight or experimental evaluation from blood or urine. Nonquantitative properties, such as the name of the favorite color of the subject, will be excluded.

Some potential biomarkers recorded in step 3 will not be useful for determination. The remaining steps of this step collect a set of "candidate biomarkers" from step 3 of the latent biomarkers. Each candidate biomarker will be selected because there is information from previous studies / knowledge that the biomarker is a potentially useful discriminator and quantitative evidence from training sample data. In each step, the biomarkers selected as candidates are removed from the list of potential biomarkers and transferred to the set of candidate biomarkers. The reason for removing a selected candidate biomarker from the list of potential biomarkers is that once the biomarker has been selected as a candidate, there is no reason to consider it, and the list is already made. At the end of the process, all unselected potential biomarkers will be removed from the following considerations, and only candidate biomarkers will be used for further analysis.

Step 4: Initialize the candidate biomarker set by including any potential biomarker that is believed to be relevant to the particular biological condition based on previous research and experience.

The purpose of this step is to use previous information on potentially important biomarkers for specific biological conditions. For example, if a particular biological condition has coronary heart disease (CHD) at a particular time, previous studies have shown that the values of serum cholesterol, systolic blood pressure, glucose intolerance, or tobacco smoking (only a few) suggest CHD. It is related to the onset of and indicates that should be replicated from the list of potential biomarkers to the list of candidate biomarkers.

Any reliable informative material or "experiment based guess" may be different depending on the selection of a subset of biomarkers known or believed to be related to a particular biological condition. Although the nature of the initially selected biomarker is not critical to determining the nature of the subset finally selected for use in the determination, it is finally determined by this system as having the greatest statistical significance to predict a particular biological condition. Initial selection of the biomarkers made will help to provide faster convergence into an empirically determined subset. In other words, the faster the initial selection is made by experience, the faster the convergence.

Step 5: Add any latent biomarker "statistically important" correlated with the "known" important biomarkers from step 4 to the candidate biomarker set.

The data obtained from the training sample is used to calculate the correlation coefficient between each previously identified candidate biomarker ("known important" biomarker) and each potential biomarker. Any statistically valid correlation coefficient can be used.

The purpose of identifying biomarkers is for good discrimination. The interrelationship of "known important" biomarkers would be a better discrimination than "known important" biomarkers themselves. At the very least, the correlation of known important biomarkers should be included in the initial analysis.

If a particular biological condition is actually defined as the value of one or more biomarkers (eg, hypertension), the defined biomarker will be a "known important" biomarker and will be moved to the list of candidate biomarkers in step 4. Correlation of defined biomarkers will be transferred to the list of candidate biomarkers at this step.

"Statistical significance" is used herein only as a means for determining between a "probably important" correlation and a "probably not important" correlation. In a preferred embodiment, conventional p values will be calculated for the correlation between potential biomarkers and candidate biomarkers. If p is smaller than any particular value, for example p <0.05 or p <0.01, the potential biomarkers are transferred to the candidate biomarker list.

Step 6; A logistic regression model for each latent biomarker is applied using the biological conditions specified as the dependent variable (Y) and the binary indicator variables for age and latent biomarker as the independent variable (X). Each latent biomarker "statistically important" in the logistic regression model is added to the candidate biomarker list.

The purpose of this step is to consider these (linear) effects of age and then select those potential biomarkers as candidate biomarkers related to the probability of having a particular biological condition. The logistic model represents the probability of having a particular biological condition as a function of the value of the latent biomarker in relation to the age of the subject.

The biomarker is selected (or not selected) based on the limit p value for the slope of the biomarker in the logistic regression model. In accordance with the above correlation, "statistical importance" is used herein only as a means for determining between "probably important" and "probably not important" determinations. The value of p will be computed for the slope of the potential biomarker, if p is less than any particular value, eg, p <0.05 or p <0.01, the potential biomarker will be transferred to the candidate biomarker list.

Step 7: Each time series assigned potential biomarker using a general linear mixed model ("MixMod") to assess whether the time series of values of the biomarker is related to the acquisition of specific biological conditions Evaluate. Each potential biomarker with statistically significant time series trends will be transferred to the candidate biomarker list.

The purpose of this step is to identify biomarkers rather than proceeding previously with candidate biomarker states having a time series tendency related to the probability that the biomarkers will have a particular biological condition.

In a typical embodiment of the present invention, each model will be provided as follows. The dependent variable (Y) in MixMod contains the time series of potential biomarkers. Independent variables (X) for fixed effects may include (1) binary indicator variables for specific biological conditions, (2) time from some appropriate event, age such as number of visits, or other relevant time-series metameters, (3) Correlation between binary indicator variables and time series metameters for specific biological conditions. The random effect portion of the model includes a random subject increase for the concept of random slope for the population regression line and in some cases the time series metameters. When two or more random effects are included, the covariance matrix of the random effects is generally unstructured. Age or other related time series meters are included in the model for the same reasons as in step 6.

If the coefficient corresponding to any X variable but not age is statistically significant, the potential biomarker is moved to the candidate biomarker list. Matters on statistical significance in step 6 are available herein.

In steps 4-7, each biomarker with historical or quantitative evidence that all potential biomarkers have been examined and is useful as a discriminator is moved to the candidate biomarker list.

Step II. The candidate biomarkers are reduced to a set of selectable biomarkers with discriminant power, and mixed model estimation of covariance structure and prediction values is performed.

background; Conventional discriminant assays include two populations, population D (i = 1) and population It is common to need a relatively accurate estimate of the covariance matrix Σ _i and the mean vector μ _i of the variance of the biomarker of (i = 2). μ _i is estimated as a simple sample mean (vector), Σ _i is estimated as a simple sample covariance matrix, and does not allow adjustment of the mean for important accompanying variables (or covariates), and facilitates repeated measurements from the same subject. It does not include. In addition, conventional discriminant analysis methods are typically based on a "case-by-case" process, where if a subject has any missing data, all data in that subject is deleted from the analysis.

Mean Vector μ ₁ and Variance Matrix Given an estimate of ₁ and a biomarker (and associated data) for the subject in vector Y, the usual discriminant function ( ₁ = ₂ is a linear function, _{One 2, the} quadratic function) is Y, μ ₁ , μ ₂ , ₁ and Evaluated from ₂ . The only information specific to the special subject is in vector Y.

The mixed model process, which accounts for most of stage II, is μ ₁ , μ ₂ , ₁ and _The general process is improved by using a general linear mixed model (MixMod) for both models (the discriminant function uses model estimates of these factors rather than the usual simple non-model estimates). This MixMod process has the following important improvements over conventional discriminant analysis:

Factors are estimated using the following mixed model:

Use all available data, i.e. not use casewise deletion,

Estimated variance matrix Support distributed adjustment of the estimated expected value (μ ₁₎ with corresponding adjustment of the _first and,

A hybrid model that supports the use of repeated measures (eg from yearly visits) from the same subject.

This MixMod process adds to the estimate of the population mean μ ₁ , which can improve the discriminant performance of the discriminant function, or uses individual random effects and model-based estimates of "BLUP" (Best Linear Unbiased Predictor) instead of the population mean.

Outline of the Phase II Process

As a result of step I, each candidate biomarker will have temporal or quantitative evidence useful as a discriminator. However, there is a substantial correlation between candidate biomarkers. Thus, biomarkers, by themselves, have real discriminating power but will not contribute substantially when used in combination with other biomarkers. In addition, the scale of the biomarker can vary widely.

The purpose of phase II of the subject process is

(1) reconvert the biomarker values so that the standard deviation of all reconverted biomarkers is about the same magnitude (0 <standard deviation ≤ 1),

(2) reduce the list of possible long terms of the candidate biomarkers to a few "Select Biomarkers", each of which substantially contributes to the discriminant power of the set,

(3) determine the structure of the expected value of the vector Y of the (reconverted) biomarker value using a linear model of the form E [Y] = Xβ, estimate the vector β of the unknown factor,

(4) Using a model of the form ZΔZ ′ + V to determine the structure of the variance matrix of the vector Y of the (reconverted) biomarker values, estimate the unknown variance factors in the matrices Δ and V,

(5) Estimate the random subject effect vector d _ik and calculate the predicted value vector Y _ki ^(P) of the k th subject even if the subject is derived from the i th specific biological condition group (i = 1 corresponds to the group D). I = 2 is a group Corresponding to).

In a representative embodiment of the present invention, step 1 of this step is performed once to reconvert the biomarker data and arrange the data into one data vector (or one variable in the dataset). Steps 2 and 3 are executed repeatedly until a set of selection biomarkers is selected and the estimates listed above are calculated. Step 4 redefines the mixed model and factor estimates to be used for discrimination by selecting the fit model for the variance matrix.

Step 1: Create a dataset in which one variable "RespScal" contains the conversion values (including time series measurements) of all candidate biomarkers from all subjects

Conversion is performed separately for each biomarker. Each biomarker value is divided by the sample standard deviation of that biomarker. Thus, the standard deviation of the converted values of each biomarker is 1.00. In an exemplary embodiment of the invention, one variable of the biomarker will be designated as "RespScal", which is an abbreviation of "Response-Scaled". The sample standard deviation of RespScal is also approximately 1.00. This conversion facilitates the procedure of an iterative process in subsequent mixed model calculations.

Step 1 is executed only once. First, all candidate biomarkers have data in RespScal and are considered members of the selected biomarker set. Critical biomarkers will be removed from the selected biomarkers in steps 2 and 3.

Step 2: Create a general linear mixed model (MixMod) using the following specifications: Obtain estimates of the factor matrices β, Δ, and V. Assuming the subject is in each particular biological condition group i = 1, 2, obtain an estimate of the random subject effect d _ik of each subject and the predicted values Y _ik ^(min) and Y _ik ^(avg) of each subject.

In a representative embodiment of the invention, the following is a detailed specification of MixMod:

Dependent (Y) variables: RespScal;

Independent (X) variable and its coefficient (β):

“Biological condition state” indicator variable (classification variable) for the state of a particular biological condition; The biological condition state is 1 if the corresponding element of Y contains information about the subject from population D, and 0 otherwise.

Biomarker indicator variables (classification variables);

Biological condition state x indicator variable (classification variable) of the biomarker;

Age (approximately year based on the overall mean age of the subject; continuous variable);

Random effect variable (Z _k ) and random coefficient effect (d _ik ):

Subject x biomarker indicator variable (part of Z _k ) and corresponding random effect (intercept increment; part of d _ik ). The random subject effect for a particular biomarker is constant over a number of visits to that subject, generating a correlation between the biomarker's repeated measurements for that subject.

Note that the model assumes E [d _ik ] = 0 and V [d _ik ] = △.

Variance matrix V _b of the vector ∈ _kb of the biomarker random error term for the k-th subject in the time series evaluation of the b-th converted candidate biomarker (∈ _kb ). This variance matrix has one row and one column for each time series evaluation of each biomarker for the kth subject. Note that the model also assumes that E [∈ _kb ] = 0.

The primary interpretation of ∈ _kbv is the “random measure error term” representing the deviation from one estimate to the other in the value of the candidate biomarker converted to the age-dependent average of subject k for that converted candidate biomarker. This interpretation _{assumes that} the value of ∈ _kbv is equally distributed and uncorrelated, that is, (k, b, v) If (k ', b'v') it is reasonable to assume that Cov (∈ _kbv , ∈ _{k'b'v '} ) = 0. When Y of the elements are classified in k (subject ID), b (biomarkers ID) and v ( "Hit" or the subjects evaluated numbers or age), a reasonable model for the V _k in a number of cases, V _k = BlockDiag (V _kb ) = BlockDiag (V _k1 , V _k2 , ...), where V _kb = λ _b I and λ _b = (∈ _kbv ), the deviation of the measurement error for the converted value of the b th candidate biomarker, the deviation is assumed to be the same for all subjects (k) and all evaluations (v).

Note that the conversion of RespScal implies that each deviation λ _b will be less than 1.00. The degree of how low the deviation is from 1.00 depends on the size of the fixed effect (high R ² causes less estimated deviation) and the size of the random effect ( Diagonal element).

The combination of Z _k , d _k , V _k = BlockDiag (V _kb ) and V _kb = λ _b I Note that we generate a highly complex extended symmetric model of _ik . Equal Deviation Factor for Group D and Group To illustrate the essence of the example when applied to all, a vector of random effects for the k th subject and the b converted biomarker is given by d _k = [d _kb ] = [d _k1 , d _k2 ,... ], (d _k ) = Δ = [δ _{bb '} ], where δ _bb' = Cov (d _kb , d _{kb '} ), b and b' can index different converted biomarkers, and Z _k is converted An indicator variable for the biomarker is included, and V _kb = λ _b I. therefore, _k = Z _k ΔZ ' _k + _k = [ _{k, b, b '} ] where _{k, b, b '} = δ _bb' J + λ _b I = variance matrix of the multiple measurements from the converted biomarker b, _{k, b, b '} = δ _bb' J = variance of the converted biomarkers b and b 'evaluated for the same case or for different cases (each element of the square matrix J is 1).

The process of building a mixed model yields estimates of:

Factors β, Δ, and V _k of the model. If the model assumes different variances over two populations of biological condition states, the model generates separate estimates of the variance factors of Δ _i and V _ik .

Expected value of data vector μ _ik of each subject (subject k is in biological condition state population i).

The expected value of the data vector μ _i'k of each subject, assuming that the subject is in a different response population (i ').

The random subject effect d _ik of each subject in the subject's actual processing population (i) and d _i'k if the subject is in a different response population (i ').

" _Predicted value" Y _ik ^(p ) of each subject in the actual processing population (i) of the subject and Y _i'k ^(p) if the subject is in a different response population (i ' ⁾ .

The variance matrix of the subject _k . If the model assumes different variances for two populations of biological condition states, then the model Generate a separate estimate of _ik .

Step 3: Delete the biomarkers with the least apparent discriminant power and reconstruct the mixed model

The biomarker to be an effective discriminator must have a large (statistically important) biological condition state x biomarker fixation effect. Conversely, large biomarker main effects are not relevant here, and large biomarker main effects that represent differences in the biomarker mean are simply increased because the biomarkers are different types of variables and have different means (on the reconverted axis). can do. In contrast, a large biological condition state × biomarker effect is [biological condition state = 0] (group The biomarker mean for) is substantially different from the biomarker mean for the [biological condition state = 1] (group D) mean for the same biomarker.

If each selected biomarker has a statistically significant biological condition state × biomarker fixation effect, step 3 is complete and proceeds to step 4. If one or more of the currently selected biomarkers have a nonstatistically important biological condition state × biomarker fixation effect, then the biomarker with the least biologically significant (maximum p-value) biological condition state × biomarker fixation effect It is removed from data vector Y and MixMod is written to the reduced data vector.

In step 3 an analog of the "backward elimination" process in the stepwise regression syntax is performed. An alternative approach is to perform an "forward selection" analogy that includes only a few clearly valid variances (biomarkers) in the data vectors and models and adds one biomarker at each subsequent step.

Step 4: determining a structure of the dispersion factor matrix △ _p and V _ik

Discriminant analysis methods are based on each biological condition state group D and For both, we use both the biomarker's expected value and the biomarker's variance matrix, some of which are evaluated in time series. It has been described previously that the list of optional biomarkers, including possible time series evaluations, has already been completed in Step 3. As mentioned above, MixMod is the variance matrix _ik = Z _ik △ _i Z ' _ik + includes the assumption that leads to the following structure for _ik , where i is the index of the biological condition state population (i = 1 for population D, and I = 2), and k indexes the subject. Further, the dispersion factor matrix Δ _i and _ik is especially _{When ik} is very large, that is, when there are multiple time series evaluations of multiple biomarkers and / or one or more biomarkers, they will have a structure that can be developed for analysis.

The purpose of step 4 is to use the variance factor matrix Δ _i for use in step III discriminant analysis and to determine the structure of _ik . The estimate of the structured large variance factor matrix will be more accurate than the estimate of the unstructured variance factor matrix. △ _i and by a more accurate estimate of _{ik ik} = Z _ik △ _i Z ' _ik + More accurate estimates of _ik can be obtained, and thus more accurate estimates of β, d _ik and Y _ik ^(p) and more accurate values of the discriminant function.

The overall structure of _ik must take into account the following types of variance / correlation:

ADB type: dispersion / correlation among different biomarkers evaluated at the same time;

ALESB type: variance / correlation in time series evaluation of a single biomarker;

BTBEL type: dispersion / correlation between two biomarkers evaluated in time series, ie, dispersion / correlation between any pair of biomarkers evaluated at different times.

In a representative embodiment of the present invention, the structures described in step 2 or extensions of these structures would be useful.

In a representative embodiment of the present invention, Tangen at the 1996 Spring Meeting of the International Biometric Scociety, Eastern North American Region, held in March 1996 in Richmond, Virginia, to develop a scatter / correlation structure for time series multivariate data. The technique disclosed in "A case study of the analysis of multivariate longitudinal data using mixed (random effects) model" by Catherine M. and Helms, Ronald W. is used. The choice of variance model usually involves the process of constructing multiple MixMods, using the same expected model, and changing the variance model. Models can be compared through log-like statistics (assuming lower normal distribution). The structure of variance can be obtained using techniques disclosed in "Model Selection Techniques for the Covariance Matrix for Incomplete Longitudinal Data," published in Statistics in Medicine, 14, 1397-1416 by Ronald W. Helms and Grady, JJ, University of North Carolina, USA. It will be compared graphically.

Step III: Compute a discriminant function using the estimated mean value and the predicted value, and compute the logical semiotic predictive value for each subject; Estimated Error Rate for Discriminant Function

background; The purpose of stage III is to have the subject "populate", ie group D or group It is "predicting" which group of people belong to.

Population D: A subpopulation of people who will acquire certain biological conditions within a certain time frame.

group : A subpopulation of people who will not acquire certain biological conditions within a certain time frame.

Subjects are classified by placing the subject into one of the following two populations.

Population PD: A group of people who are expected to acquire a particular biological condition within a particular time zone at the start of a particular time zone, that is, a person who is expected to belong to group D. These people are described as those who have a defined high probability of obtaining certain biological conditions within a certain time frame.

group : A group of people, or groups, who are expected to fail to acquire certain biological conditions within a certain time zone at the beginning of a particular time zone. A group of people predicted to belong to. These people are described as those who have a specified low probability of obtaining certain biological conditions within a certain time frame.

The second purpose is for subjects D and Is to estimate the probability of belonging to.

Techniques for achieving the first object, that is, techniques for classifying a subject into one of two populations, use a discrimination process that is a variation of conventional discriminant analysis. An estimate of the probability that a subject will be in a group of subjects that will acquire a particular biological condition is obtained from a variation of conventional logical semiotic regression using (1) a discriminant function value as the regression value and (2) a discriminant variable as the regression value. do.

As described in the background of step II, prior art discriminant analysis methodologies typically employ a mean vector μ _i and a variance matrix of the distribution of biomarkers of two populations. _Use the raw estimate of _i . Moreover, prior art discriminant analysis is based on a "case-by-case" process in which all of the subject's data is deleted from the analysis if the subject has any missing data.

The mixed model process described in step II is μ ₁ , μ ₂ , ₁ and _The general process is improved by using a general linear mixed model (MixMod) to model all of them, and the modeled estimates of these factors are used in the discriminant function instead of the usual simple unmodeled estimates. By using a mixed model, this process will yield the following significant improvement over conventional discriminant analysis. The factor is estimated using all available data. That is, the argument does not use case-by-case deletion. This process uses the estimated variance matrix It supports distributed adjustment of the expected value (μ _i) estimating a corresponding adjustment of the _i. This process then supports the use of repeated measures (eg, from yearly visits) from the same subject.

The use of a mixed model allows the process to utilize individual random effects and model-based estimates of Best Linear Unbiased Predictor ("BLUP") in addition to or in place of estimates of the population mean μ _i to increase the discriminant performance of discriminant functions. It will be more important that you can.

The format of this discrimination is identical in form to the ordinary discrimination based on multivariate normality. Some comments like this are useful:

f _i is μ _i and the evaluation by using the estimate of _i, represents the distribution of the density function of the vector Y of the determined parameters for the subjects from group i, and i = 1 for the group D, or PD, group or For i = 2;

p _i indicates the likelihood that the subject will be from group i, i = 1 for group D, and Is i = 2. The value of p _i is known by temporal data or other studies. If the value of p _i is unknown, then the proportion of subjects in the two populations will be used as the estimate of p _i .

An unknown group of subjects with a vector Y of discriminant function values will be classified as group 1 (group PD) if Ln [f ₁ (Y) / f ₂ (Y)]> Ln [p ₂ / p ₁ ] , Otherwise, Group 2 (Group Will be classified as).

In step II, two populations have the same variance matrix, i.e. ₁ = ₂ = We will decide whether we can reasonably assume cognition. In that case, this discrimination process is reduced by the use of a linear discriminant function of the form:

D (Y) = [Y-½ (μ ₁ + μ ₂ )] ' ^-1 (μ ₁ -μ ₂ )-Ln [p ₂ / p ₁ ]

μ ₁ and _i will be replaced by an "suitable" estimate, described below. D (Y) vs. 0 will be compared. In step II ₁ ≠ If determined to be ₂ , the discrimination process is reduced by the use of a second order discrimination function of the form:

Q (Y) = ½ln (| ₂ | / | ₁ |)-½ (Y-μ ₁ ) ' ₁ ^-1 (Y-μ ₁ ) + ½ (Y-μ ₂ ) ' ₂ ^-1 (Y-μ ₂ )-Ln [p ₂ / p ₁ ]

Where μ _i and ₁ is replaced by the "suitable" estimate described below. Q (Y) vs. 0 are compared.

In both cases, the "suitable" estimate is derived from the blending model process in step II and may or may not include random subject effects.

Step III Process-The steps of Step III of the process are described as follows. Assume that data is available from one or more "new" subjects, i.e., subjects whose group membership is not known and are not used in calculating the mixed model in Step II. In steps 1 and 2, one subject will be considered at a time. Some additional notations are available. I = 1 for groups D and PD, and And For i, i = 2.

Y represents a vector of discriminator values for one new subject. The element of Y is converted when RespScal is converted in step II.

X _i represents a matrix of values of the independent variables used in the final phase II mixture model, assuming that the subject is in group i (where i = 1, 2). The row of X _i corresponds to the row (element) of Y.

Z _i represents the matrix of values of the random valid variable used in the final phase II mixture model, assuming that the subject is in group i (where i = 1, 2). The row of Y _i corresponds to the row of Y.

Denotes the estimated variance matrix of the random effects from population i (i = 1, 2) in the final phase II mixed model. In many cases the mixed model is reduced to a single variance of random effects. In other words, to be.

Denotes the estimated variance matrix of the random residual term or “error term” from population i (i = 1, 2) in the final phase II mixed model. In many cases the mixed model is reduced to a single variance of random effects. In other words, to be.

Denotes the estimated variance matrix of Y from the final phase II mixed model, assuming that the new subject is from population i (i = 1, 2). In many cases the mixed model is reduced to a single variance matrix. In other words, to be.

Step 1: Classify all the subjects in the verification sample using the results from the Step II mixed model, and utilize one combination of the estimated random subject effects to determine one candidate process and "prediction" based on "estimated values". The error rate of the plurality of candidate discrimination processes consisting of the remaining candidate discrimination processes is estimated. The process with the lowest estimated error rate is selected and referred to as the "highest reliable process".

If the original learning population is divided into a "training sample" and a "validation sample", then a validation sample is used in the following cases, otherwise the training sample is used as a valid sample. If the subjects are from each population, then separately estimate the following properties for each subject in the valid sample.

Assuming the subject is from group i (i = 1, 2), it is the "estimated value" of Y.

Assuming the subject is from group i (i = 1, 2), it is an estimate of the random subject effect of the subject.

Back side ; Otherwise - silver And It will be considered as the "minimum value" or "minimum (over population) random subject effect" estimate of.

- silver And Will be considered as an "average" or "average (over population) random subject effect" estimate of.

The "prediction" of the subject when the subject is from group i (i = 1, 2) but using a "minimum" random subject effect estimate.

The "prediction" of the subject when the subject is from the group i (i = 1, 2) but using the "average" random subject effect estimate.

In the foregoing and the following, i = 1 for group D or group PD, and or Is i = 2.

Estimate Classification based on

₁ = ₂ = Is determined in Step II, μ _i To replace To Evaluate the linear discriminant function D (Y) by substituting for. If D (Y) ≥0, the subject is designated as group 1 (group PD); otherwise, the subject is group 2 (group). )

_{One If} the determination of ₂ is made in step II, μ _i To replace _i To evaluate the quadratic discrimination function Q (Y). If Q (Y) ≥0, the subject is designated as group 1 (group PD); otherwise, the subject is group 2 (group). )

"Minimum" random subject effect and prediction Classification based on

₁ = ₂ = Is determined in Step II, μ _i To replace To Evaluate the linear discriminant function D (Y) by substituting for. If D (Y) ≥0, the subject is designated as group 1 (group PD); otherwise, the subject is group 2 (group). )

_{One If} the determination of ₂ is made in step II, μ _i To replace _i To evaluate the quadratic discrimination function Q (Y). If Q (Y) ≥0, the subject is designated as group 1 (group PD); otherwise, the subject is group 2 (group). )

"Average" random subject effects and predictions Classification based on

₁ = ₂ = Is determined in Step II, μ _i To replace To Evaluate the linear discriminant function D (Y) by substituting for. If D (Y) ≥0, the subject is designated as group 1 (group PD); otherwise, the subject is group 2 (group). )

_{One If} the determination of ₂ is made in step II, μ _i To replace _i To evaluate the quadratic discrimination function Q (Y). If Q (Y) ≥0, the subject is designated as group 1 (group PD); otherwise, the subject is group 2 (group). )

After each subject in the verification sample (as described above) is classified, for each of the three processes (based on estimated or predicted values), 2 × 2 similar to the following is calculated:

The number of subjects in the verification sample tabulated by actual and classified member relationships in D Subject is classified as a member of a group PD Subject is actually a member of a collective = Number of divisions = Number of false positive classifications D = Number of false negative classifications = True number of false classifications

Furthermore, separate classifications based on estimates and classifications based on predictions:

= False positive justice ratio = false positive proportion

= False negative rate of error = false negative rate of classification

= Total error rate = ratio of false classifications

In a representative embodiment of the invention, three types of classification processes, i.e., estimates And "minimum" predictions And "average" predictions based on The classification process based on the above will be compared to determine the "apparent most reliable process". Some considerations in the selection process include:

If the false negative classification has more serious consequences than the false positive classification, then choose a process with less false negative error rate r _FN . This situation will arise, for example, if group D is a subpopulation of people suffering from myocardial infarction (“MI”) within a particular five year age group. False classification of false, which is a warning failure for a person with a high MI probability, will have more serious consequences than false definition classification, which is a warning for a low probability person with a high MI probability.

Conversely, if the false positive classification has more serious consequences than the false negative classification, choose a process with fewer false positive error rates r _FP .

If there is no prior reason to specify greater severity for false negative classification or false positive classification, choose a process with a lower total error rate r _tot .

As the most reliable course, the course chosen will be divided into two collective PDs and Used to classify as

Step 2: Use two types of logical semiotic regression to calculate an estimate of the probability that the new subject will belong to each group.

The data from the training sample is used as the independent ("X") variable of the discriminant function for each subject (D (Y) for the first order, Q (Y) for the second order) and the biological condition state (group D). Indicator variables for members within are used to create a logical semiotic regression model, which is used as a dependent ("Y") variable. The model is used to calculate for each subject an estimate of the probability that the subject belongs to group D, along with the inverse logical semiotic transformation.

In a separate calculation, the data from the training sample is used as the biomarker used in the discriminant function as an independent ("X") variable with the final mixed model variance (variable in X), and the member of the biological condition state (group D). Indicator variable for) is used as the dependent ("Y") variable. In addition to obtaining general logical semiotic regression model estimates, a model is used to calculate the estimated probability that the subject will belong to group D for each subject with an inverse logical semiotic transformation. When time series data is used, a model is used to estimate the probability that a subject belongs to group D at the end of a particular period. A generalized estimating equation method with a logical semiotic link function can be used to obtain correlation among multiple binominal outcomes from a subject.

The prediction probabilities from these two models can provide a target interpretation of the discriminant function values.

Although the subject algorithm is a preferred implementation for determining the discriminant function to be used in a subject, this algorithm is provided for the purpose of illustrating the preferred implementation of the invention and is not intended to limit the invention to the steps or substeps of the algorithm. For example, in the field of discriminant analysis techniques there are other types of discriminant functions, such as, for example, "optimal discriminant", other types of regression, such as nonlinear mixed models, which are included in the technical idea of the present invention. Will be.

The invention will now be described with reference to specific examples, materials, devices and processing steps in an exemplary sense. In particular, the present invention will not be limited to statistical methods, materials, conditions, treatment factors, devices, and the like.

Examples of Preferred Embodiments

The accompanying tables and figures show the results of examples of data analysis using the methods and procedures of the present invention.

The data used as the basis for this example is obtained from a database containing patients whose sickle cell data are obtained based on the year. Some patients have data obtained from three consecutive visits. However, because a patient cannot usually be forced to participate year after year, the database includes a large number of patients whose data is only available for one or two year visits. Database information used herein includes demo data, clinical chemistry data and hematology data.

Particular biological conditions of the subject (“disease” or “infection”) in this example are the occurrence of painful ferns requiring hospitalization. At an annual visit, the subject is asked if they have recently suffered a painful crunch that requires hospitalization (and is determined by examining the records). Each subject recorded as hospitalized for painful attrition at any visit (any year) becomes a member of the "onset" group (Group D), and all other subjects are grouped. Become a member of

Whenever a subject has recently experienced a painful expenditure that requires hospitalization, all data collected after hospitalization in that year or another year is excluded from the analysis. This aids in the process to be used in the event of death or the occurrence of chronic or incurable disease. Variables that record the population D member relationship of the subject (eg, whether they are present or infected) are referred to as “disease state” variables.

The following is an example of a statistical analysis using sickle cell data. For confidentiality reasons, the data used in this example are artificially written and not derived from actual research or actual subjects. But. This data is similar to the data that can be obtained in the study of real subjects.

Step Ⅰ. Selection of biomarkers for confirmation and consideration of evaluation methods

Step 1: Choose a methodology to estimate the error rate of the process

Step 2: Select the "Training Sample", which is a subset of the test population, and the Complementary subset, "Effective Sample," to be used for the statistical analysis that leads to the discriminant / probability estimation process.

For this example, a training sample / effective sample method is selected. Patients are randomly assigned to one of two samples. The training sample is used to write the discriminant function, and the test sample is used to evaluate the accuracy of the discriminant function.

The training sample contains information from 641 "yearly" assessments from 481 subjects, that is, information from 1.3 annual assessments per subject. However, not all biomarkers are evaluated even if the subject visits. For example, only 88 values of Direct Bilirubin (variable L_DBILI) are available from only 80 subjects.

Step 3: Compile a List of Potential Biomarkers as Potential Determinants

In this case, blood pressure, all available demo data, clinical chemistry data and hematology data are used as potential discriminations. Potential biomarkers are listed in Table 2.

Step 4: Initiate the establishment of candidate biomarkers by including any potential biomarkers believed to be involved in specific biological conditions throughout previous research and development

In this example, platelet dimensions are taken as "unknown" biomarkers for disease states.

Step 5: Add to the list of candidate biomarkers any potential biomarkers that are "statistically important" correlated with "unknown significant" biomarkers from step 4.

A biomarker is selected that correlates to platelet dimensions, which is the "unknown important" biomarker from step 2. An overview of these correlations is shown in Table 3, denoted "correlation W / platelet dimensions". The "p" row represents the p-value for correlation with platelet dimensions. Biomarkers are selected based on gain p-values for Poisson product-moment correlation coefficients. In this example, p <0.01 is required for selection. The column “p <cv” indicates the biomarker that is a candidate biomarker as a result of the “significant” correlation with platelet dimensions by the presence of the word “YES”.

Step 6: A logical semiotic model is created for each biomarker with disease status as the dependent (Y) variable and age and biomarker combination as the independent (X) variable. In this case, for each biomarker, the logical semiotic model evaluated the age of the subject and how well the probability of inpatient care for painful fern was explained by that biomarker. Roughly speaking, the regression coefficient or slope of a biomarker in a logical semiotic regression will be approximately zero if there is no relationship between the biomarker and the probability that the subject will acquire a particular biological condition, indicating that there is a nonzero slope relationship. . An overview of the logical semiotic regression results is shown in the column heading "Logical semiotic regression" in Table 3. The "p" column represents the p-value for the regression coefficient of the biomarker. Biomarkers are selected based on the extra p-value for the slope of the biomarkers of the logical semiotic regression model. In this example p <0.01 is required for selection. The column "p <cv" indicates the biomarker that becomes a candidate biomarker as a result of the "significant" logical semiotic regression coefficient by the word "YES". Some of these biomarkers are significantly associated with platelet dimensions and become candidate biomarkers before logical semiotic regression is calculated.

Step 7: Potential biomarkers evaluated for each time series using a general linear mixed model ("MixMod") to assess whether the time series trend in the value of the biomarker is related to the acquisition of specific biological conditions Rate it

A mixed model is created for each biomarker, with the time series of the biomarker as the dependent (Y) variable and the age, disease state, and number of visits × disease state as independent (X) variables. Class ", and that coefficient is an increment for the intercept. In contrast, age is a continuous variable whose coefficient is the slope). The random effect portion of the mixed model contains the correlation between time series measurements from the same subject. The model can change the number of visits (time series estimates) from subject to subject.

The biomarker will be selected if the disease state "main effect" or one of the three visits x disease state action subvectors is statistically significantly different from 0 (P <0.01). A prominent disease state "main effect" is that the average of the biomarker values for the subjects in group D is grouped. Different from the mean for the subject within. 3 The prominent subvector of the number of visits × disease state action coefficients is that the temporal trend in the biomarker values for the subjects in group D It will indicate that it is different than the temporal trend for the subject within. In both cases (significant main effect or action), the result will indicate that the biomarker is a potentially available determination and should be moved to the candidate biomarker list. The results from the mixed model are shown in the heading of the “mixed model” in Table 3. Separate results are shown for the main effects and actions in a format similar to the results from correlation and logical semiotic regression.

Upon completion of steps 4-7, all potential biomarkers are examined and each biomarker with temporal or quantitative evidence of utilization as a discrimination is moved to the list of hobo biomarkers. Candidate biomarkers are indicated by the word "YES" in the column of heading "Selected" in Table 3.

Step II. Reduce candidate biomarkers to a set of selective biomarkers with discriminant verification power, and perform mixed model estimation of variance structure and predicted values

Step 1: Create a data set with the variable "RespScal" containing the conversion values (including time series measurements) of all candidate biomarkers from all subjects

This step is performed for example, and the result is not displayed. However, all of the values of the whole different biomarkers are arranged in one row vector Y, which may include a number of elements.

Step 2: Build a general linear mixed model (MixMod) using the detailed specifications listed below, obtain estimates of the factor matrices β, Δ, and V, and assume that the subject is in each specific biological condition group i = 1, 2. Assume a random subject effect d _ik of each subject and the predicted values Y _ik ^(min) and Y _ik ^(avg) of each subject

Step 3: Reconstruct the Mixed Model by Removing the Biomarkers with the Minimum Discriminant Verification Power

Steps 2 and 3 are repeated until all biomarkers in the mixed model are statistically significant. To simplify the description of the example, only the final result of the iteration over steps 2 and 3 is described. Steps 2 and 3 reduce the number of biomarkers to 15 with age as the fixed effect variance.

General information about the example mixed model is provided in Table 4. Data from 481 patients with up to three visits for each patient are available. Note that many observations are not used in the analysis. Artificial observations are generated using missing Y values to allow the software to calculate the required predictions. Artificial observations with missing Y values do not affect the estimation of the factor or the prediction of the random subject effect.

Table 5 provides estimates of the fixed effects from the mixed model. The p-value for each biomarker (eg, the p-value for "L_BUN") is the p-value for the hypothesis test that the mean value of this biomarker is equal to the overall mean averaged over the entire biomarker. do. The fact that these p-values are significant is not very important and one can predict that the mean value of one biomarker will be different from the mean value of another biomarker.

In Table 5, the p-value for each "biomarker X GROUP IA" action (eg, the p-value for "ALBUMIN X GROUP IA") is the mean of the biomarkers for group D. It is the p-value for the test of the hypothesis that it is significantly different from the mean value of the biomarker for. A prominent value (eg p <0.05) indicates that the biomarker should be a good judgment. All of the actions in the final model represented by Table 5 are statistically significant (total p≤0.05). Age remains in the model even if the p-value is not significant.

The subject, biomarker, disease state (“population”) and visit observations and predictions for subject 447 are shown in Table 6. This subject is a collection (With no "RESPSCAL" for rows where "group D?" = No. "Group D?" = YES), but has predictions for all groups. This subject does not have data for the biomarker MCH or MCHC for the second visit, but has a model based prediction for Visit 2 MCH and MCHC for that subject.

The method implemented in steps 2 and 3 is an analog of the "backward delete" process of the stepwise regression syntax. Another method is to first perform a "backward selection" analogue that includes only two (or very few) significant validity determinations (biomarkers) in the model, and add one biomarker at each subsequent step. have.

Step 4: Determine the structures of the dispersion factor matrices Δ and V _ik

As mentioned above, The overall structure of _ik must consider three types of variance / correlation:

ADB type: dispersion / correlation among different biomarkers evaluated at the same time

ALESB type: dispersion / correlation during time series evaluation of a single biomarker

BTBEL type: dispersion / correlation between two biomarkers evaluated in time series, ie dispersion / correlation between any pair of biomarkers evaluated at different times

In the example, the following structure is finally obtained:

Cohort D and The same random effect variance factor matrix for all, i.e. Δ ₁ = Δ ₂ = Δ, and

Δ has a complex symmetry, δ _ii = 0.6669, δ _ij = 0.0097 for i ≠ j,

Cohort D and Same for all, matrix with complex symmetry, i.e. V _ii = 0.3267, V _ij = 0.0151 for i ≠ j ADB type distribution within.

This scatter structure is reasonable for certain sickle cell data.

△ and The estimate of is shown in Table 7. The evaluation estimate of the variance matrix Δ of the random subject effect is shown at the top of the table. Rows and columns correspond to the 15 biomarkers used in this model, and are shown in columns.

Variance Matrix within Visit Error in Subject The estimate of is shown at the bottom of the table. Like △, rows and columns are the model Corresponds to the 15 biomarkers used in. Has a complex symmetric structure that is reasonable for the conversion data.

Step III: Compute the discriminant function using the estimated mean and the predicted value, calculate the logical semiotic predictive value for each subject, and estimate the error rate for the discriminant function

Step 1: Classify all the subjects in the verification sample using the results from the Step II mixed model, and based on the "estimated value" and a number of candidates based on the "predicted value" utilizing various combinations of estimated random subject effects. Estimate the error rate of the discrimination process. The process with the lowest estimated error rate becomes the selected process and is referred to as "apparently the most reliable process".

The invention is applied using mixed model results for sickle cell data. Variance factor matrix of group D and group Since it is modeled as the same for, each discrimination becomes a linear discrimination. Each determination is applied to a subject within a training sample (used here as a verification sample) so that each subject is a collective PD and Will belong to one of the

The evaluation of the subject linear discriminant function based on the estimates is shown in Table 8. Population with disease state = "no" 100 subjects (56%) out of 179 subjects in the population And 79 subjects (44%) are incorrectly classified as collective PD. Of the 262 subjects in population D with disease state = "YES", 188 subjects (72%) were correctly classified as group PDs by discrimination, and 74 subjects (28%) were incorrectly grouped. Classified as Of the total 441 subjects, 288 subjects (65%) are classified correctly, and 35% are classified incorrectly.

Table 9 shows the evaluation of the subject linear discriminant function based on prediction values using the minimum random subject effect. Table 9 is similar to Table 8. Predictive Discrimination It shows some discriminant improvement in the population, but results in somewhat poor results in group D. Overall the error rate is about the same.

The preceding paragraph and the classification / misclassification statistics in Tables 8 and 9 are biased. That is, since training samples are used to obtain and evaluate the discriminant function, the table provides better misclassification estimates than are likely to occur. Evaluation of the discriminant function using the evaluation sample will provide an estimate of the deviation of the misclassification rate. Resampling techniques such as jackknifeing or bootstrapping can produce estimates that are less convenient while using data from training samples.

Step 2: Use two types of logical semiotic regression to calculate an estimate of the probability that a new subject will belong to each group

Two types of logical semiotic regression are fitted to the training sample data for each of the discriminant functions. In both logical semiotic regressions, the disease state indicator is a dependent ("Y") variable. In the first logical semiotic regression, the value of the discriminant function based on the estimate is used as an independent ("X") variable. In the second logical semiotic regression, the value of the discriminant function based on the prediction is used as an independent ("X") variable. In the third logical semiotic regression, the biomarker used in the discriminant function is included as an independent ("X") variable with the variance used in the fixed effect portion of the mixed model, and the disease state indicator is a dependent ("Y") variable. Included as. The estimates from the logical semiotic model are used to calculate the estimated probability that for each subject this subject belongs to the disease (disease state "YES") population. The results of the logical semiotic regression calculations are not shown in the table.

1 shows group D (solid line) and group The experimental discriminant function ("EDF") of the linear discriminant function value (based on the estimated value) for (dotted line) is shown. To build the graph, the data for the subject is sorted by increasing the value of D (Y) in the disease state population and within the population. Data points are floated in that sequence. The EDF starts at 0 (before the data in the first subject is floated) and is incremented by 1 / n for each subject, where n is the number of subjects in the population. Thus, EDF rises separately from 0 to 1 for each population. In Figure 1, the EDF for population D is population The fact that the D is shifted to the left of the EDF for Indicating a trend with a lower score. Approximately 72% of population D is zero (group PD and Group with a value less than We can see that we have about 44% of the EDF value of the subject on the left side of zero. The slope of the EDF of the population adjacent to the vertical line at LDF = 0 indicates that many subjects are not easy to classify in the "boundary line". If additional years in the next year are available The number of subjects (of these data) will suffer pain in subsequent years and will be "inverted" for group D.

Group D (solid line) and Group The experimental distribution function (“EDF”) of the minimum random subject linear discriminant function value for (dotted line) is shown in FIG. 2. Results and interpretations are similar to those in FIG. 1. However, the collective EDF line is more inclined near LDF = 0 in FIG. 2 than in FIG. 1, indicating the fact that many subjects are at the borderline.

These figures, like the statistics described above, indicate that the dispersion process efficiently classifies subjects that must be hospitalized due to pain, but are less efficient for subgroups that will not be hospitalized because the data available in this example is limited. Table 2 shows the description of potential biomarkers for sickle cell data, and Tables 3a and 3b show the selection of candidate biomarkers from a list of potential biomarkers using Step I, Steps 5-7, Correlation, Logistic Regression, and Mixed Models. Table 4 shows mixed model information, overall model characteristics, and Tables 5 a and 5b show estimates of the fixed effect coefficients and related statistics, and Tables 6a, 6b, 6c, and 6d show predictions for subject 447 and Table 7a and 7b show the estimates of the covariance matrix, Table 8 shows the evaluation of the discriminant process using the estimates, and Table 9 shows the evaluation of the discriminant process using the predicted values.

"p <cv" stands for "p-value is less than the threshold", where cv = 0.01.

Blank in p <cv means "NO".

All p-values are rounded to 2.

The number of subjects in the verification sample tabulated by actual and classified member relationships in D Subject is classified as a member of a group Yes PDNo Subject is actually a member of a collective no D Example

r _tot = 153/441 = 35%

The number of subjects in the verification sample tabulated by actual and classified member relationships in D Subject is classified as a member of a group no PD Example Subject is actually a member of a collective no D Example

r _tot = 155/441 = 35%

Claims (25)

In computer-based systems for predicting the future health of an individual, (a) a subpopulation D of a member is identified as having acquired a particular biological condition within a particular time period or age interval, and the subpopulation of said member A computer comprising a processor comprising a database of time-series obtained biomarker values from individual members of a test population, wherein s is identified as not obtaining specific biological conditions within a particular time period or age interval; (b) (1) subpopulations D and Selecting a subset of biomarkers for discrimination between members belonging to the group based on a distribution of biomarker values of individual members of the test population; (2) (i) a subpopulation PD having a certain high probability of acquiring a particular biological condition within a particular time period or age interval for a member of the test population or a predetermined biological condition within a particular time period or age interval. Subpopulation with low probability Classified as belonging to, or (ii) quantitatively estimate the probability of acquiring a specific biological condition within a specific time period or age interval for each member of the test population; And a computer program comprising using the distribution of the selected biomarker to develop a statistical process that can be used to. The method of claim 1, wherein the statistical process is performed with subpopulation D and And a discriminant function utilizing an estimated mean vector and an estimated variance matrix of the distribution of biomarker values within. The computer-based system of claim 2, wherein the estimate of the factor of the distribution of the selected biomarkers is obtained by constructing a general linear mixed model for the biomarkers from the test population. The method of claim 2, wherein (a) the estimated mean vector is modeled as a vector value function of the expected factor or variance, or (b) A computer-based system in which the estimated variance matrix is modeled as a variance factor or matrix value function of variance values. The computer-based system of claim 4, wherein the estimate of the factor of the distribution of the selected biomarkers is obtained by constructing a general linear mixed model for biomarker data from the test population. 5. The computer-based system of claim 4 wherein the estimated mean vector or probability comprises an estimate of a member from which the real value or probability of the random subject effect vector for the member being classified is estimated. The computer-based system of claim 6, wherein the estimate of the factor of the distribution of the selected biomarkers is obtained by constructing a general linear mixed model for biomarker data from the test population. In computer-based systems for predicting the future health of an individual, (a) a subpopulation D of a member is identified as having acquired a particular biological condition within a particular time period or age interval, and the subpopulation of said member A processor comprising a processor comprising a database of biomarker values from individual members of a test population, wherein is identified as having not acquired a particular biological condition within a particular time period or age interval; (b) (1) subpopulations D and Selecting a subset of biomarkers for discrimination between members belonging to the group based on a distribution of biomarker values of individual members of the test population; (2) (i) a subpopulation PD having a certain high probability of acquiring a particular biological condition within a particular time period or age interval for a member of the test population or a predetermined biological condition within a particular time period or age interval. Subpopulation with low probability Classified as belonging to, or, (ii) quantitatively estimate the probability of acquiring a specific biological condition within a specific time period or age interval for each member of the test population; Using the distribution of the selected biomarker to develop a statistical process that can be used to generate a statistical process, the statistical process comprising And a computer program comprising a discriminant function utilizing an estimated mean vector of the distribution of biomarker values within and an estimated variance matrix. The computer-based system of claim 8, wherein the estimate of the factor of the distribution of the selected biomarkers is obtained by constructing a general linear mixture model for biomarker data from the test population. 10. The method of claim 9, wherein (a) the estimated mean vector is modeled as a vector value function of an expected factor or variance, or (b) A computer-based system in which the estimated variance matrix is modeled as a variance factor or matrix value function of variance values. 11. The computer-based system of claim 10, wherein the estimated mean vector or probability comprises an estimate of a member of which the real value or probability of the random subject effect vector for the member being classified is estimated. In the method of predicting an individual's health, Collecting from the individual a plurality of biomarker values from which the at least one biomarker value is obtained by physically measuring the biomarker value; (i) classify the individual as having a certain high probability of obtaining a particular biological condition within a particular time period or age interval or having a predetermined low probability of obtaining a specific biological condition within a particular time period or age interval. Or, (ii) applying a statistical process to the plurality of biomarker values to quantitatively estimate the probability of obtaining a particular biological condition within a particular time period or age interval for the individual. , The statistical process is, (1) the subpopulation D of the member is identified as having acquired a particular biological condition within a particular time period or age interval, and the subpopulation of the member Collecting a database of time-series obtained biomarker values from individual members of the test population, wherein is identified as not obtaining specific biological conditions within a particular time period or age interval; (2) subpopulation D and from the biomarker Selecting a subset of biomarkers for discrimination between members belonging to the group based on a distribution of biomarker values of individual members of the test population; (3) using the distribution of the selected biomarkers to develop the statistical process. The method of claim 12, wherein at least one of the biomarker values is obtained from a biological sample. The method of claim 13, wherein the biological sample is a serum sample or urine sample. In computer-based systems that predict the future health of an individual, (a) a computer having a processor containing a plurality of biomarker values from an individual; (b) collecting from the individual a plurality of biomarker values from which at least one biomarker value is obtained by physically measuring the biomarker value; (i) the individual as a person with a certain high probability of acquiring a specific biological condition within a particular time period or age interval, or as a person with a certain low probability of acquiring a specific biological condition within a specific time period or age interval. Or (ii) applying a statistical process to the plurality of biomarker values to quantitatively estimate the probability of obtaining a particular biological condition within a particular time period or age interval for the individual. Computer program, The statistical process is, (1) the subpopulation D of the member is identified as having acquired a particular biological condition within a particular time period or age interval, and the subpopulation of the member Collecting a database of time-series obtained biomarker values from individual members of the test population, wherein is identified as not obtaining specific biological conditions within a particular time period or age interval; (2) subpopulation D and from the biomarker Selecting a subset of biomarkers for discrimination between members belonging to the group based on a distribution of biomarker values of individual members of the test population; (3) using the distribution of the selected biomarkers to develop the statistical process. The computer-based system of claim 15, wherein the plurality of biomarker values from the individual is a biomarker value obtained in time series. The computer-based system of claim 15, wherein the particular biological condition is death due to certain potential causes of death within a certain time period or age interval. The computer-based system of claim 15, wherein the specific biological condition is a specific morbidity within a specific time period or age interval. 16. The computer based system of claim 15, wherein the particular time period is a minimum two year period. The computer-based system of claim 15, wherein the particular time period is at least three years. In a method of determining an individual's future risk of death due to certain potential causes of death, Collecting from the individual a plurality of biomarker values from which the at least one biomarker value is obtained by physically measuring the biomarker value; The individual has a certain high probability of dying within a particular time period or age interval due to any of the potential causes of death that account for at least 60% of all deaths in the test population over a particular time period or age interval. And applying a statistical process to the plurality of biomarker values to determine whether they are classified as. In the method of judging individual's health proof, Collecting from the individual a plurality of biomarker values from which the at least one biomarker value is obtained by physically measuring the biomarker value; There is a certain high probability that the individual will not die within a particular time period or age interval due to any of the potential causes of death that account for at least 60% of all deaths in the test population over a particular time period or age interval. And applying a statistical process to the plurality of biomarker values to determine whether they are classified as having. In computer-based systems that determine the risk of future death for an individual due to certain potential causes of death, (a) a computer having a processor comprising a plurality of biomarker values from an individual; (b) any high probability that the individual will die within a particular time period or age interval due to any of the potential causes of death that account for at least 60% of all deaths in the test population over a particular time period or age interval. And a computer program comprising applying a statistical process to the plurality of biomarker values to determine whether they are classified as having a probability. In a computer-based system for determining the evidence of individual health, (a) a computer having a processor comprising a plurality of biomarker values from an individual; (b) a predetermined period of time in which the individual will not die within a particular time period or age interval due to any of the potential causes of death that account for at least 60% of all deaths in the test population over a particular time period or age interval. And a computer program comprising applying a statistical process to the plurality of biomarker values to determine whether they are classified as having a high probability. In a device for determining the risk of an individual's future health problems, (a) a storage device for storing a plurality of biomarker values from an individual; (b) is connected to the storage device, 1) programmed to receive the plurality of biomarker values from the storage device, 2) (i) a subpopulation PD having a certain high probability of acquiring a particular biological condition within a particular time period or age interval, or a predetermined low probability of acquiring a specific biological condition within a specific time period or age interval. Subpopulation with Programming to apply a statistical process to the biomarker value to classify as belonging to, or (ii) to quantitatively estimate the probability of obtaining a particular biological condition within a particular time period or age interval for the individual. Processor, The statistical process is, (1) the subpopulation D of the member is identified as having acquired a particular biological condition within a particular time period or age interval, and the subpopulation of the member Collects a database of time-series obtained biomarker values from individual members of the test population identified as having not acquired a particular biological condition within a particular time period or age interval; (2) subpopulation D and from the biomarker Selecting a subset of biomarkers for discrimination between members belonging to the group based on the distribution of biomarker values of individual members of the test population; (3) a determination device based on using the distribution of the selected biomarker to develop the statistical process.